Resultados de la aplicación de la encuesta a los docentes de la carrera

In this appendix, I present and discuss additional empirical exercises to confirm that the correlations presented in the paper are robust.

The first exercise pertain the cutoff in the number of analysts’ forecast re- quired for an observation to be included in the data. In the man text, I consider a cutoff of 5, but I claim this choice does not affect the results. To show that this is the case, Table 2.3 presents the fixed effect regression results for cutoffs ranging from 2 to 7.43

43

Evidently, two is the minimum number of forecasts needed to be able to actually compute a coefficient of variation. Robustness to even higher cutoffs (in particular, ten) is presented in Table

Table 2.3: Robustness Check (1): different cutoffs

(1) (2) (3) (4) (5) (6)

Cutoff 2 Cutoff 3 Cutoff 4 Cutoff 5 Cutoff 6 Cutoff 7 L.CV forecasts 0.0303∗∗∗ 0.0238∗∗ 0.0236∗∗ 0.0215∗∗ 0.0249∗∗ 0.0240∗∗ (3.63) (3.09) (3.10) (2.72) (2.86) (2.70) L.Total Assets 0.00947∗ 0.0140∗∗ 0.0131∗∗ 0.0133∗∗ 0.0127∗ 0.0136∗ (2.20) (3.27) (2.91) (2.83) (2.51) (2.50) L.Profitability -0.0476 -0.138∗∗∗ -0.143∗∗∗ -0.156∗∗∗ -0.158∗∗∗ -0.171∗∗∗ (-1.70) (-9.43) (-8.59) (-8.31) (-8.28) (-8.16) L.Book-to-Market 0.00528∗ 0.00221 0.00171 0.00147 0.00196 -0.00185 (2.31) (0.90) (0.51) (0.40) (0.46) (-0.34) L.Intangibles 0.0322 0.0286 0.0279 0.0163 0.0156 0.00911 (1.48) (1.31) (1.24) (0.71) (0.66) (0.36) L.Industry Leverage 0.386∗∗∗ 0.348∗∗∗ 0.340∗∗∗ 0.339∗∗∗ 0.316∗∗∗ 0.322∗∗∗ (8.03) (7.32) (6.84) (6.53) (5.88) (5.82) Constant 0.282∗∗∗ 0.272∗∗∗ 0.283∗∗∗ 0.285∗∗∗ 0.304∗∗∗ 0.302∗∗∗ (6.54) (6.35) (6.24) (5.87) (5.86) (5.36)

Time FE Yes Yes Yes Yes Yes Yes

Firm FE Yes Yes Yes Yes Yes Yes

Observations 35263 32512 29472 26465 23686 21150 AdjustedR2 0.842 0.845 0.846 0.845 0.848 0.848 t statistics in parentheses. ∗p <0.05,∗∗p <0.01,∗∗∗p <0.001

Sources: Compustat merged with CRSP (annual), IBES (detail, adjusted for stock splits). Notes: all independent variables are lagged by one year. Stnd. errors clustered at the firm level.

From now onwards, by ‘Usual Controls’ I shall refer to those included in the regressions of Table 2.3.

The second set of robustness checks, presented in Table 2.4, studies how the results change with different measures of analysts’ forecast dispersion. Column (1) reports the benchmark estimate using the coefficient of variation (it is equivalent to column (4) of Table 2.2). Column (2) clarifies the importance of normalising the standard deviation by the mean: without the normalisation the significance is lost. Column (3) and (4) do the same replacing CV with MAD (the median absolute deviation from the mean forecast). Similar results attain. Finally, column (5) shows that one could also use directly the number of analysts following the firm in a given year. As expected, the number is negatively correlated with leverage, suggesting that the higher the number of analysts following a firm, the lower its subsequent leverage ratio.

Table 2.4: Robustness Check (2): different independent variables

(1) (2) (3) (4) (5)

LT/AT LT/AT LT/AT LT/AT LT/AT

L.CV forecasts 0.0249∗∗ (2.86) L.STDEV 0.0000123 (0.03) L.MAD forecasts 0.0487∗∗∗ (3.31) L.MAD*MEAN 0.00103 (0.67) L.Estimates -0.00120∗∗ (-3.20)

Time FE Yes Yes Yes Yes Yes

Firm FE Yes Yes Yes Yes Yes

Usual Controls Yes Yes Yes Yes Yes

Observations 23686 23686 23686 23686 23686

AdjustedR2 0.848 0.848 0.848 0.848 0.848

t statistics in parentheses. ∗p <0.05,∗∗p <0.01,∗∗∗p <0.001

Sources: Compustat merged with CRSP (annual), IBES (detail, adjusted for stock splits). Notes: all independent variables are lagged by one year.

Standard errors are clustered at the firm level. 2.2 in the main text.

The third series of robustness checks is presented in Table 2.5. It considers the effects on the estimates of changing the definition of leverage. In particular, column (1) presents again the estimates shown in the main text, where leverage is defined as in Welch [2011], to equal the ratio of Total Liabilities (LT) over Total Assets (AT). Column (2) replaces AT with the market value of assets (AM = MEQ + LT). The coefficient of interest is positive but looses a one degree of significance. Column (3) shows what happens when leverage is defined as the ratio of Total debt (DT) – defined as the sum of Debt in Current Liabilities (DLC) and Long Term Debt (DLTT) – over the book value of assets. The result is similar to that of column (2). Finally, column (4) shows what happens when leverage is defined as DT/AM. The coefficient looses significance altogether. Columns (5)-(7) repeat the exercise of substituting LT/AT with alternative measures of leverage for the independent variable MAD. Similar results attain.

Table 2.5: Robustness Check (3): different dependent variables

(1) (2) (3) (4) (5) (6) (7)

LT/AT LT/AM DT/AT DT/AM LT/AM DT/AT DT/AM L.CV forecasts 0.0249∗∗ 0.0218∗ 0.0186∗∗ 0.0112

(2.86) (2.50) (2.67) (1.76)

L.MAD forecasts 0.0457∗∗ 0.0356∗∗ 0.0232 (2.76) (2.63) (1.82)

Time FE Yes Yes Yes Yes Yes Yes Yes

Firm FE Yes Yes Yes Yes Yes Yes Yes

Usual Controls Yes Yes Yes Yes Yes Yes Yes Observations 23686 23686 23646 23646 23686 23646 23646 AdjustedR2 0.848 0.884 0.794 0.813 0.884 0.794 0.813 t statistics in parentheses. ∗p <0.05,∗∗p <0.01,∗∗∗p <0.001

Sources: Compustat merged with CRSP (annual), IBES (detail, adjusted for stock splits). Notes: all independent variables are lagged by one year. Stnd. errors clustered at the firm level.

Finally, Table 2.6 explores the leads and lags structure of the data. Although CV is serially correlated, the Table shows that the results are stronger when CV is assumed to precede leverage than the other way around. Of course, the results do not rule out reverse causality, and a statistically causal analysis is still required in future work.

Table 2.6: Robustness Check (4): lags and leads

(1) (2) (3) (4) (5) (6)

LT/AT LT/AT LT/AT LT/AT LT/AT LT/AT L3.CV forecasts 0.0295∗∗ (2.77) L2.CV forecasts 0.0266∗∗ (3.06) L.CV forecasts 0.0249∗∗ (2.86) CV forecasts 0.0650∗∗∗ (6.84) F.CV forecasts 0.0176∗ (2.06) F2.CV forecasts 0.00637 (0.77)

Time FE Yes Yes Yes Yes Yes Yes

Firm FE Yes Yes Yes Yes Yes Yes

Usual Controls Yes Yes Yes Yes Yes Yes Observations 18597 20994 23686 23686 20568 17811 AdjustedR2 _{0.860 0.855 0.848 0.849 0.855 0.858} t statistics in parentheses. ∗p <0.05,∗∗p <0.01,∗∗∗p <0.001

Sources: Compustat merged with CRSP (annual), IBES (detail, adjusted for stock splits). Notes: all independent variables are lagged by one year. Stnd. errors clustered at the firm level.